Re: Brief freezes with -current as of early last week

On Mon, Sep 20, 2004, Nik Clayton wrote:
> After updating from -current built on 2004/04/27 to -current built on
> 2004/09/13, I'm seeing some annoying 'freezes'.  I'd not seen these with
> the earlier -current.
> 
> To complicate matters, right around the time I did the update I also
> switched from a configuration with two SCSI disks and an IDE disk, to a
> configuration with two IDE disks (each SCSI disk, bought at the same
> time, from the same supplier, died within 4 days of each other --
> sheesh).
> 
> So, I guess what I might be seeing is plain old I/O contention --
> although I'm surprised that it freezes the box.  According to
> 'atacontrol mode 0', both IDE devices are using UDMA100, which is the
> maximum the motherboard (dual 1GHz PIII) supports).
> 
> The symptoms typically manifest themselves right after kicking off a
> portupgrade command.  The system will (sort-of) freeze.  If I'm in X,
> the mouse cursor stops responding, there's no response to the keyboard.
> However, if I happen to have top(1) running in an xterm, the top(1)
> display keeps updating.  This lasts (wall-clock) 10 to 15 seconds, and
> then the system becomes responsive again.

This sounds like an I/O hang.  I've been having similar problems
related to WRITE_DMA and READ_DMA timeouts from the ata driver in
the last week or so, as discussed last week on this list.  Are you
seeing these messages from the kernel?  Julian suggested disabling
preemption, which I haven't tried yet, but other people report
limited success.